Principal Designer at Microsoft AI. Building https://xitter.cf/t.co/sxWcklVrid + AmyraAI. Product design, interaction, motion, and taste.
Seattle, WA
Joined April 2009
- Tweets388
- Following282
- Followers237
- Likes529
Been exploring the virtual try-on landscape and built a prototype that feels like a meaningful shift.
Most try-on today is still image-generation based, give the system a photo + garment, generate a result, then repeat. This explores real-time try-on directly in the camera. You can move naturally, switch products instantly, and see the clothing stay with you continuously.
What’s especially cool is the fidelity, fabric wrinkles and folds, drape, shadows, highlights, and even the room’s lighting carry through into the clothing. It starts to feel less like a preview and more like you’re actually wearing it.
That changes the loop from select → generate → inspect to browse → wear → swipe → compare.
Still early, but this feels like try-on starting to become part of browsing itself.
#tryon #design #shopping
Building Oddflick taught me that discovery isn't a search problem. People rarely know the title they want. They know the mood, the face, or the five seconds that made them stop scrolling.
The most interesting Elyx idea isn't AI generation. It's putting design files beside the code, where Git, agents, and humans can share the same source of truth.
xitter.cf/pieteromvlee/status/20…
So excited to finally be able to share what we've been working on: Elyx, a new design tool for a new age. elyx.design
The foldable iPhone will be judged less by how it opens and more by whether people stop thinking about the fact that it folds.
AI has created a strange new kind of FOMO: you're producing more than ever and still feel behind. The bottleneck moved from making things to deciding what deserves your attention.
Design find: this status picker by @nitishkmrk makes presence feel tactile.
One dot expands into Message, Focus, and Mute; selection collapses it back into one color-coded state.
Tiny surface, clear feedback, zero modal energy.
AI can generate 40 interface directions before lunch.
Amazing: we can explore more.
Less amazing: someone still has to say “hmm, not quite” 39 times.
The bottleneck didn’t disappear. It became taste, judgment, and an unreasonable number of open tabs.
The best interface motion doesn’t decorate a state change. It explains it.
What moved?
Where did it go?
What can I do next?
If the transition answers those questions, it reduces cognitive load. If it only looks impressive, it’s visual noise.
Streaming gave us everything to watch and somehow made 100 years of cinema invisible.
I built Oddflick to bring it back: swipe through classic film scenes, tap when one catches you, and watch the whole movie.
Free on iPhone.
I solo-built the entire app, and this time I’m proud of the engineering I put into Oddflick - The feed learns from what you watch and what you skip, so it gets more like you over time — while keeping enough randomness that it never narrows down to one thing. Every film is re-encoded and broken into small pieces, then served from whichever server is nearest you — which is what lets you jump into the middle of a 90-minute movie instead of downloading the whole thing first.
Varun Varshney retweeted
We’ve been working on voice models that feel genuinely expressive. Curious what you think! Try the latest out in the MAI playground: playground.microsoft.ai/